How DARPA's R&D model brings AI advancements to the enterprise

DARPA was commissioned in 1958 by The Advanced Research Projects Agency “ARPA” to develop military technology for the nation. It was renamed in 1972. The mission of the research agency was to create solutions to some of the US’ most challenging defense and national security challenges.

However, DARPA’s contributions to the non-military world are extensive, including funding and directing the development of the Moderna vaccine. In fact, at a recent event for their new book, retired CIA Chief of Disguise Jonna Mendez revealed how the CIA’s Directorate of Operations leaned on DARPA during the Cold War.

But DARPA’s unique nature allows all but the most sensitive discoveries to find their way into the public sector. Consider just five DARPA Inventions that changed the world:

• The Computer Mouse (1964)
• The Internet (1969)
• GPS (1983) .
• “Siri” (2002) (originally called "Calo")
• Drones (1988)

How is it funded:

The FY2022 budget is $3.528 billion. Nearly all of DARPA's funding falls under the budget categories of basic research, applied research, and advanced technology development of the Department of Defense. DOD refers to financing the science and technology (S&T) budget under these categories. 

DARPA comprises approximately 220 government employees in six technical offices, including nearly 100 program managers, overseeing about 250 research and development programs. This is a relatively small staff for such a large budget, the result that most of the DARPA funding is awarded in grants. 

In addition, grant recipients have the opportunity to commercialize their work. 

How it matters to the enterprise. For almost five decades, DARPA has led the application of rule-based and statistical-learning-based AI technologies. Today DARPA funds a wide range of activities from basic research to technology development, including systems that can acquire knowledge through contextual and explanatory models, an effort referred to as the “Third Wave” AI technologies. In 2018 DARPA embarked on a $2 billion program, the “AI Next” campaign. Not all programs are public knowledge, but some that have been announced are:

• Upgrading DoD business processes, such as or accrediting software systems security clearance, securing for deployment.
• Impressive advances in the reliability of AI systems (DARPA is leading the world in methodology and technique for explainability in Machine Learning (ML).
• Productizing the security and resiliency of ML and AI technologies; 
• Vastly reducing power consumption
• Discovering the  AI algorithms for “explainability” and common sense reasoning.

We hear a great deal from commercial developers of AI, but DARPA has an ambitious plan for the next few years:

• Real-time analysis of dangerous cyber attacks
• Detection of deep fakes
• Human language
• Biomedical innovation
• Control of prosthetics

Another area where DARPA is contracting is deploying more robust AI with 5G wireless networks, a top priority worldwide to enable the Internet of Things. Other projects DARPA is working on today that have a direct effect on the enterprise beyond the Defense Department:

• Neural Nets. The science of neural nets is well known and applied across military and civilian systems, but how they evolve and become reliable is still not understood. 
• Neural nets and reinforcement learning are being applied in many places, but they’re not robust yet. Training and retraining them, for example, is still a bit of black art. 
• One of the problems in building more reliable and predictable AI is the lack of commonly agreed methodologies for AI systems engineering. 
• One of DARPA’s projects is creating a reliable AI system to enable flight control systems to adapt to conditions and keep the plane within a feasible fight control envelope. According to DARPA’s website, the Aircrew Labor In-Cockpit Automation System (ALIAS) program “envisions a tailorable, drop-in, removable kit that would promote the addition of high levels of automation into existing aircraft, enabling operation with a reduced onboard crew.” It adds that ALIAS “aims to support execution of an entire mission from takeoff to landing, even in the face of contingency events such as aircraft system failures.”

AI technologies have already been shown to be valuable for missions as diverse as space-based imagery analysis, cyberattack warning, supply chain logistics, and analysis of microbiologic systems.

My take

Formal verification — with the notion of flying in or trusting your life to decisions being made by a system autonomously — today, I wouldn’t do it without having humans involved, without having controllers. And make no mistake, DARPA’s primary mission is still the Department of Defense. Seventy years ago, the most advanced tank in the world had little application to the consumer market. Still, technology is today primarily digital, and DARPA is more likely to find commercial and medical applications, if we don’t blow each other up first. 

My favorite DARPA project at the moment is DARPA's program to reduce mosquito attraction and bites, which is moving into its second phase.